Agent / Applied AI Engineer

Chatbots are yesterday. We're building the AI-first ERP. ERP runs how real companies make and ship physical things. It has traditionally been a maze of screens, tabs, and fields that takes months to learn. We think that era is ending. We're building a product surface where the user describes what they want and an agent does it — surfacing the right action, the right insight, the right next step. This is not a chat window bolted onto the old UI. It's a composable, AI-first experience that makes the complexity disappear. We have a clear vision and roadmap, and we give people real room to shape it. Why Rootstock, why now: Rootstock brings proven depth in ERP and manufacturing — real products, real customers, decades of hard-won domain knowledge. We're putting that depth behind a clear blueprint for becoming AI Native, with active development underway and the funding to see it through. You'd build alongside a deep bench of senior engineers who have shipped real ERP at scale. Iron sharpens iron. The role in one line: You design and build the agents, tools, and AI-native interactions that turn a complex ERP into something that feels like it's working for the user. What you'll build • Agents with real authority over real workflows. Not Q&A toys — agents that take action inside genuine business processes like orders, inventory, and financials, safely and correctly. • Typed, intent-routed tools over MCP. Our agents run on MCP servers hosted outside Salesforce, with typed tools and grounded retrieval. ERP is an enormous surface area and you can't stuff it all into a single MCP. You'll solve real problems in scale and performance while building on high-value customer use cases. • A composable, AI-first UI. We're building a typed component library designed so both humans and models produce correct UI from it. You'll help shape it and assemble surfaces that present actions and insights, not just forms and tables. • The evaluation harness behind all of it. Agents are only as good as your ability to measure them. You'll build and own the evals that tell us whether an experience is actually getting better. What we're looking for • Hands-on experience building with LLMs in production: agents, tool-calling, MCP or equivalent, retrieval and grounding, prompt and context engineering. You've shipped something real and watched it behave in the wild. • Strong engineering fundamentals and genuine product sense. You can feel when an interaction is right. Competence in Java, JavaScript, or a comparable language matters more than React or TypeScript expertise. • A serious point of view on agent design: how to scope a tool, how to keep an agent grounded, how to fail safely, and how to evaluate. • Curiosity about the domain. Deep ERP knowledge isn't required. The ability to turn intimidating complexity into a clean surface is. You'll thrive here if... • You experiment with AI for fun. Have you built an agent, a bot, or some weird AI tool just to see what it could do? That's exactly the instinct we want. • You love rapid iteration — build a proof of concept, prove or kill the theory, and move on. You're comfortable with non-determinism and you debug it with evals, not vibes. • You're independent, self-directed, and hungry to leave a mark. You'll help shape what AI-first ERP becomes, not just maintain someone else's spec. • You have taste. You can tell the difference between an AI feature that's a gimmick and one that genuinely removes work for the user. How we work We do genuinely new things, and new things fail often. We expect to chase ten ideas and have seven hit dead ends. Nobody gets punished for a smart bet that didn't pay off — the job is to run the experiment, learn from it, share what you found, and keep moving. We write up our dead ends openly, because a well-understood failure shapes the next decision as much as a win does, often more. We care as much about raising the whole team as we do about shipping the next feature. People here treat helping a teammate level up on AI as the fun part, not a tax on their time. No egos. We push each other to get sharper. What we'll go deep on in interviews Models and their tradeoffs, the agentic experiences you've built, how you keep an agent grounded and safe, how you evaluate quality, and the hardest problems you've hit shipping AI to actual users. Come with opinions and examples. The basics • Location: Remote, US-based • Visa Sponsorship: This position requires authorization to work in the US; we are unable to provide visa sponsorship, including H-1B. • Compensation: Base salary, annual bonus, and a long-term incentive plan (LTIP). We set the level by experience — show us what you've built. Your work has a direct line to the LTIP upside. • Tooling: Claude license, full access to Salesforce's AI tooling, and approved API spend as needed. We don't want budget standing between you and an experiment. • How to apply: Send us the agent, the demo, the repo, the thing you built th